Could you set up a special Hudson build that would only run the
Solaris-failing test on Solaris, and would use a skunk branch rather
than the trunk?
Given that, I could debug on Solaris-Hudson, by editing in logging
and/or printouts. I use System.err printouts when I want to ask a
specialized question that is probably only needed for that step of that
debug effort, and I can't use Eclipse to get the data.
It would be slower than debugging on a machine under my direct control,
but not that much.
I did get some failures from the QA test of my TxnManagerImpl change, so
I need to investigate, and may need to modify the fix.
Patricia
Jonathan Costers wrote:
Hi Patricia
Great work again.
I'm going to look into creating a separate Hudson build to run only on
Solaris nodes today. Likewise, we'll have another one running on Ubuntu
nodes only.
The former should fail consistently, the latter should pass consistently.
On first sight, the issue on Solaris seems to have to do with multicasting
somehow, but haven't been able to spend much time to investigate.
Best
Jonathan
2010/10/31 Patricia Shanahan <[email protected]>
I've found a bug fix to com.sun.jini.mahalo.TxnManagerImpl that makes the
entire javaspace category pass. I've started a full QA run with the fix, but
that will take about a day. If anything fails as a result of the
TxnManagerImpl change I'll have to debug that, but until I get a failure
there is nothing more to do on javaspace.
There is arguably a problem in the txnmanager test category because it did
not detect the bug. Maybe I should add the txnmanager category to a couple
of the javaspace tests that use transactions.
Time pick my next bug hunt - I can start something else while the QA test
is running. Any opinions?
I could take a look at some of the skipped tests to see why they are
skipped. Maybe some of them would tell us about real bugs if we ran them.
There is also the Solaris-only bug. I can build a Solaris VirtualBox and see
if it reproduces. If not, I may need to learn how to run things on a Hudson
Solaris in order to investigate.
Patricia
On 10/20/2010 1:44 PM, Jonathan Costers wrote:
Great job Patricia!
My vote would go to the "javaspace" test category.
Last time I ran that one (250 or so tests IIRC) I got 16 failures.
If we can get these cleared up, I believe we have a solid test base in
place
to start validating some new developments and experiments.
Looks like we are really getting some momentum here, I like it a lot.
Thanks to all for your hard work.
Jonathan
2010/10/20 Patricia Shanahan<[email protected]>
On 10/19/2010 4:08 PM, Patricia Shanahan wrote:
I propose modifying TxnManagerImpl to make it match the interface
declaration, and allow an abort to be retried. This may break other
tests, if they are assuming the behavior that TxnManagerImpl
implemented.
I'm doing a full QA test, including txnmanager. Although the test is
still
running, all of the txnmanager tests, including GetStateTest, have
passed.
Those tests are the most likely to notice the change.
If the rest of the QA test is clean when it finishes, I'll check in the
fix
and we can add txnmanager to the default category list.
Any votes on my next bug hunt? For example, are there bug reports in Jira
that really need to be fixed before the next release?
Patricia
